Deep Learning & Natural Language Processing: Transformers

Following our demographic example, we might choose a demographic subgroup and have an algorithm suggest small changes or even full rewrites of existing text to make it score more highly for a given demographic. Recall that based upon the vast amount of clickthrough rate data we have, we can build models that can take a given text and predict its CTR (click-through rate) for a given demographic. Having this model we can then “run it in reverse” and back fill by starting with a given text and moving itgenerating modified contentthat our initial model tells us is more likely to received a high CTR from the target demographic.

Given a document, completely rewrite the document in a different format. E.g., translate a long movie review of Inception in the style of The New Yorker into a short movie review of the same movie in the style of Wired. Obviously this tends to be someone what easier going from long to short documents.